Real-time foul detection in football matches using machine learning techniques

Abstract

As a widely popular sport worldwide, football necessitates precise and consistent decision-making to uphold fair game-play. It has become essential to automate and optimize certain aspects of the game for fairness and efficiency. Foul detection stands as one of the most challenging and contentious areas where this could be applied. This paper presents an approach for real-time foul detection in football matches using advanced machine-learning techniques. Our research focuses on developing and validating a machine learning-based model that uses video feed data, position coordinates and historical match data to detect fouls in real-time. Faster R-CNN, YOLOv5, YOLOv8 and YOLO-NAS like SOTA machine learning models have been used for this research due to their higher processing speed and accuracy at real time object detection workings. For the detection of foul, machine learning models YOLOv5, YOLOv8, YOLO-NAS and Fast R-CNN have shown an accuracy of about 96%, 97%, 94% and 90% respectively. The potential impact of this system extends beyond football, offering a framework that could be adapted to automate decisionmaking in various sports, thereby ushering in a new era in sports technology.